Complex -> multifactorial traits Phenotype -> categorical ...

Multivariateanalysisofcomplexphenotypes

Complex->multifactorialtraitsPhenotype->categorical&quantitativefeatures

Analysis->association&modeling

NGS(Comprehensiveevaluationallgenevariants)GWAS(mainlycommonallele,MAF>0.05)

Univariategeneassociationanalysis

Distributedriskassociatedloci (regulatoryvariants) Raredamaginglossoffunctionmutations(functionalvariants)

Multifactorialassociations

Marginaladditiveeffect ofseveralvariants Thephenotypedependsongenotypecombinations

Checkforthepresenceofinteractionsdeviatingfromapureadditivebehavior

POLYGENICRISKSCORE EPISTASIS

Epistasis analysis reveals associations between gene variants and bipolar disorder

Twocomputational/statisticalissues:

• Curseofdimensionality

• Lowstatisticalpowerfollowingmultipletestadjustment

Imputation(commonalleles)basedonreferencehaplotypes

GenotypesSNPs Haplotypereference Boostedsignal

Genotypematrix(incomparisontoreference) Eachsampleisa

mosaicofhaplotypes

Imputedgenotypematrix

GenotypesSNPs Haplotypereference Boostedsignal

Genotypematrix(incomparisontoreference) Eachsampleisa

mosaicofhaplotypes

Imputedgenotypematrix

HaplotypePhasing

• Allowcross-platformscomparison:1. Meta-analysis2. PRSdifferentGWAS

Minimac3

PRSanalysisandlogistic/linearregression

Case/controlGivenabaseassociationandtargetgenotype(afterimputation)

• IdentifycommonSNPs

• Clumptofilterforindependentsignal

• ComputePRSoverarangeofp-valuethreshold

• Identifybestfittingmodel

Permutationsarepossibletocorrectformultipletesting(nonindependenttest)

Metabolicdata:matrixoflevelsinblood

Commonpolygenicvariationcontributestoriskofschizophreniaandbipolardisorder(PMID19571811)

RESULTS:

• Thereisanadditivepolygeniccomponentinpsychiatricphenotypes

• Thereisasharedgeneticscomponentbetweenschizophreniaandbipolardisorder

• Thereisnorelevantoverlapbetweenthegeneticscomponentofschizophreniaandnon-psychiatrydiseases

• TheeffectismainlyduetoSNPsinannotatedgeneregions(ratherthanintergenicregions)

WhatwecanlearnfromPRS

CommonpolygenicvariationenhancesriskpredictionforAlzheimer'sdisease. PMID26490334

PRSandprediction

Arapidmethodforcombinedanalysisofcommonandrarevariantsatthelevelofaregion,gene,orpathwayPMID22888262

PRSincludingrarevariants

StandardPRS Weighbasedonallelefrequency

Weigh/frequencyrelationParabola(minimumatq=0.5,f=intercept) Screenoverdifferentfvalues

Accumulationofminorallelesandriskpredictioninschizophrenia PMID:28916820

PathwayspecificPRS

DisproportionateContributionsofSelectGenomicCompartmentsandCellTypestoGeneticRiskforCoronaryArteryDiseasePMID:26509271

PRSLocation

Increased”association”withinregulatoryregionslinkedwithgeneexpression

Agene-basedassociationmethodformappingtraitsusingreferencetranscriptomedata PMID:26258848

Estimationofgeneexpressionalterationstartingfromgenotypedata

l Reducemultipletestburden(from105/106 variantsto103/104 genes)

l Simplifiedmeta-analysisofgene-basedresults

l Multipletissuescanbeevaluatedusingareferencetranscriptomedataset

Heritabilityandprediction

A comprehensive simulation study onclassification of RNA-Seq data 28832679

Multiclass cancer classification based on gene expression comparison.24918456

MachinelearningclassifierstartingfromimputedgeneexpressionDifferentlyfromgeneticvariantsdata,quantitativegene“score”

canbemoreeasilyintegratedinmachinelearningpipeline